Skip to main content

CA SB 53

Transparency in Frontier Artificial Intelligence Act (TFAIA)

First US frontier AI transparency law. Requires large AI developers (>$500M revenue) to publish governance frameworks, submit quarterly risk reports, and report critical safety incidents. Applies to models trained with >10^26 FLOP.

Jurisdiction

California

US-CA

Enacted

Sep 29, 2025

Effective

Jan 1, 2026

Enforcement

California Attorney General (exclusive authority)

Signed September 29, 2025; core provisions effective January 1, 2026

Who Must Comply

This law applies to:

  • Frontier developers (models trained with >10^26 FLOP)
  • Large frontier developers ($500M+ annual revenue)

Capability triggers:

computeThreshold (required)
Required Increases applicability

Who bears obligations:

Exemptions

Federal Government Activities

high confidence

Lawful federal government activity exempt

Conditions:

  • • Federal government entity

Non-AI Information Sources

high confidence

Harm from publicly accessible information not derived from AI

Conditions:

  • • Information available from non-AI sources

Safety Provisions

  • Frontier AI Framework publication (governance, risk mitigation, cybersecurity, incident response)
  • Quarterly catastrophic risk assessment reports to Office of Emergency Services
  • Critical safety incident reporting within 15 days (24 hours if imminent danger)
  • Transparency reports before deploying new frontier models
  • Whistleblower protections with anonymous reporting channels
  • Prohibition on materially false statements about catastrophic risk

Compliance Timeline

Jan 1, 2026

Core publication, reporting, truthfulness, and whistleblower obligations take effect

Jan 1, 2027

OES anonymized reporting; CalCompute framework report due to Legislature

Enforcement

Enforced by

California Attorney General (exclusive authority)

Penalties

$1M

Max fine: $1,000,000

Civil penalty up to $1,000,000 per violation based on severity

Quick Facts

Binding
Yes
Mental Health Focus
No
Child Safety Focus
No
Algorithmic Scope
Yes

Why It Matters

First-in-nation frontier AI transparency law. Only applies to very large models (10^26 FLOP threshold) and large developers ($500M+). Sets precedent for federal AI legislation. Focus on catastrophic risk (50+ deaths or $1B+ damage), not consumer protection.

Recent Developments

Signed by Governor Newsom September 29, 2025. Successor to vetoed SB 1047 - removed kill switch, mandatory audits, cloud provider liability. Creates CalCompute public computing cluster (unfunded, contingent on appropriation).

What You Need to Comply

Large developers must: publish Frontier AI Framework with governance/risk/cybersecurity practices; submit quarterly risk reports to OES; report critical incidents within 15 days; maintain whistleblower channels.

NOPE can help

Cite This

APA

California. (2025). Transparency in Frontier Artificial Intelligence Act (TFAIA). Retrieved from https://nope.net/regs/us-ca-sb53

BibTeX

@misc{us_ca_sb53,
  title = {Transparency in Frontier Artificial Intelligence Act (TFAIA)},
  author = {California},
  year = {2025},
  url = {https://nope.net/regs/us-ca-sb53}
}

Related Regulations

Enacted US-CA AI Safety

CA SB 942

Requires large GenAI providers (1M+ monthly users) to provide free AI detection tools, embed latent disclosures (watermarks/metadata) in AI-generated content, and offer optional manifest (visible) disclosures to users.

In Effect US-CA AI Safety

CA CPPA ADMT

California Privacy Protection Agency regulations establishing consumer rights and business obligations for Automated Decision-Making Technology (ADMT) that makes significant decisions including healthcare. Requires pre-use notice, opt-out rights, access rights, appeal rights, and risk assessments.

Enacted US-NY AI Safety

NY RAISE Act

Requires large AI developers of frontier models operating in New York to create safety protocols, report critical incidents within 72 hours, conduct annual reviews, and undergo independent audits. Creates dedicated DFS office funded by developer fees.

Enacted US-VT Child Protection

VT AADC

Vermont design code structured to be more litigation-resistant: focuses on data processing harms rather than content-based restrictions. AG rulemaking authority begins July 2025.

In Effect US-NE Child Protection

NE AADC

Nebraska design code blending privacy-by-design with engagement constraints (feeds, notifications, time limits) aimed at reducing compulsive use.

Enacted US-AR Child Protection

AR HB 1958

Requires all Arkansas public entities to create AI policies with mandatory human-in-the-loop for final decisions. Covers state departments, schools, and political subdivisions.